Previous results indicating that radial maze performance in animals with mediodorsal thalamic lesions is deficient cannot exclude the possibility that these impairments are due to altered motor mechanisms (response biases). The present study sought to eliminate this potentially confounding variable by using a procedure which tests memory for serial position. This procedure involved experimenter-controlled arm entry (number and order) on a radial arm maze. Following this sequence, animals were presented with one previously entered arm along with an arm not yet visited on that trial. Avoidance of the previously entered arm constituted memory for the prior sequence. Thus, this task represents a form of a win-shift or nonmatching-to-sample task. Seven animals were given ibotenic acid lesions of the mediodorsal nucleus and nine others were given sham operations; 2 weeks later testing in the above procedure was conducted. Results indicated that, although control subjects could differentiate between entered and unentered arms without difficulty, animals with lesions were unable to exhibit this distinction. Much of their memory for arms previously entered in a sequence was at chance level, regardless of the placement of the tested arm in the sequence. Some tendency toward increased errors with longer sequences of arm entries was noted. This may indicate that animals with lesions were susceptible to proactive interference from previous choices. Regardless, even without the opportunity to develop or exhibit response biases, animals lesioned in the mediodorsal nucleus were unable to perform this win-shift task reliably. Thus, discrimination among maze arms is impaired after lesion of the mediodorsal nucleus and this impairment is independent of the motor response patterns which emerge during solution of a conventional radial maze task.